1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
/*
* IOchain - Distribute a chain of dependant IO events amoung threads.
*
* This file is part of Bitshuffle
* Author: Kiyoshi Masui <kiyo@physics.ubc.ca>
* Website: http://www.github.com/kiyo-masui/bitshuffle
* Created: 2014
*
* See LICENSE file for details about copyright and rights to use.
*
*
* Header File
*
* Similar in concept to a queue. Each task includes reading an input
* and writing output, but the location of the input/output (the pointers)
* depend on the previous item in the chain.
*
* This is designed for parallelizing blocked compression/decompression IO,
* where the destination of a compressed block depends on the compressed size
* of all previous blocks.
*
* Implemented with OpenMP locks.
*
*
* Usage
* -----
* - Call `ioc_init` in serial block.
* - Each thread should create a local variable *size_t this_iter* and
* pass its address to all function calls. Its value will be set
* inside the functions and is used to identify the thread.
* - Each thread must call each of the `ioc_get*` and `ioc_set*` methods
* exactly once per iteration, starting with `ioc_get_in` and ending
* with `ioc_set_next_out`.
* - The order (`ioc_get_in`, `ioc_set_next_in`, *work*, `ioc_get_out`,
* `ioc_set_next_out`, *work*) is most efficient.
* - Have each thread call `ioc_end_pop`.
* - `ioc_get_in` is blocked until the previous entry's
* `ioc_set_next_in` is called.
* - `ioc_get_out` is blocked until the previous entry's
* `ioc_set_next_out` is called.
* - There are no blocks on the very first iteration.
* - Call `ioc_destroy` in serial block.
* - Safe for num_threads >= IOC_SIZE (but less efficient).
*
*/
typedef struct ioc_ptr_and_lock ptr_and_lock;
typedef struct ioc_chain ioc_chain;
void ;
void ;
void * ;
void ;
void * ;
void ;
// IOCHAIN_H